This AI Expert Says Knowledge Work is Vanishing. What Does It Mean for Managers?

Most knowledge work, Jay Hack predicts, is about to move into agents.

Hack runs AI at ClickUp. In a recent podcast interview, he said that "most people who are paying attention believe" a future dominated by agents is coming. He has a front-row seat to that shift.

He’s also describing a near-term shift many companies still are not prepared to manage.

If more execution moves into agents, the worker above that execution gets a redesigned job. They still own the call, the quality bar, the handoff, the miss, and the cleanup.

They spend less of the day dragging tasks across the line by hand. They spend more of it on direction, judgment, review, and context.

That sounds productive. It also threatens to break the way most companies evaluate performance.

Management has long relied on watching the work. Drafts in the doc. Tasks moving across the board. Comments, edits, and the visible texture of effort that lets a manager guess who is carrying the load.

When more human work transitions to agents, what happens when that visibility disappears?

"Agents are limited at this point basically by the context they have access to," Hack says.

"If you give it nothing, it's only going to be a chatbot. If you give it everything, it can do what a human can do and much much more."

He is framing that as an engineering constraint. Inside a company, it lands as a labor shift.

Someone has to know which notes matter, which meeting changed the direction, which dependencies are still alive, which customer detail cannot get dropped, and which steps still need human eyes every time.

Part of the job now is deciding what context gets pushed into a machine, what stays with a person, and what gets discarded.

That is already starting to sort workers into different camps.

In our recent article, the memorable detail was that one operator had built 37 agents and regularly used 15 to 20 of them. The more important detail was what happened to his role. He stopped carrying every repeated task alone and started directing a system around himself.

Hack says the pattern is larger than one operator's experiment.

The workers who pull ahead in the next two years will be the ones who learn how to shape context, route work cleanly, and catch weak machine output before it spreads. Their peers may see the result without seeing what produced it.

That is how status shifts begin before leadership has a name for them.

Hack says companies are getting closer to being able to "assign a much larger segment of work to an agent" and have it come back as "a finished artifact."

That’s a bigger claim than it first sounds.

Once work comes back as an artifact, the human role on top of it changes shape. The day becomes a series of completed objects to review, with the process behind each compressed into something a manager cannot fully see.

The gap is already visible in the data.

In WorkAfter's upcoming report on AI tools, 65% of surveyed knowledge workers said the prompt-and-revise effort required to get usable output is disproportionate to the quality of the outputs.

The effort did not disappear. It moved into a less visible layer of steering, rewriting, checking, and stitching together machine output.

Companies still reward visible effort more easily than invisible system design. That creates a risk. The ambitious people who learn to direct machine labor will start looking faster and calmer before the org chart knows what to call that skill.

Hack says the new model of human work is moving toward direction, coordination, and judgment, the very things that are hardest to see and evaluate.

The outside evidence points in the same direction.

In a recent dataset shared by ClickUp, agent-involved messages rose from 24.3% of total chat volume to 37.1% in the first 4 months of 2026.

That metric groups together both deep delegations and lighter one-shot interactions, so it is best read as directional rather than exact. The direction is still clear. Human-to-agent interaction is becoming normal work behavior. 

Gallup's AI Indicator found that employees whose managers actively support AI use are 1.7 times as likely to use it regularly and 7.4 times as likely to say it helps them do what they do best every day.

That leaves managers with a serious set of questions.

Those questions now sit in the middle of performance review, promotion, and trust.

The wrong response is to try to preserve the old system by measuring visible signs of AI use. Prompt counts, dashboard badges, agentic-work rollups, adoption milestones, all of that tries to recreate a world that is already slipping.

The better response is to rebuild what evaluation measures.

Watch for “silent” operators. One early signal that someone has learned to direct machine labor well is that they stop looking busy in the old way.

Replace status meetings with artifact walkthroughs. Ask what they directed, what they overrode, what the system got wrong, and what they almost missed.

Audit review templates. Anything that still rewards response speed, hours logged, ticket volume, meeting presence, or generic "goes above and beyond" language is measuring activity an AI-fluent worker may already be producing less of.

Near the end of the interview, Hack says, "The people who can get the most context that represents what they're trying to accomplish in one place are going to be the best able to use agents."

He was describing workers.

He may have been describing the managers above them too.

Because the companies that notice this shift early will have a better shot at recognizing their strongest performers before the work, and the value, move too far out of sight.

Want to learn more? Watch the full interview with Jay Hack on YouTube.